Ethics-First Lightweight AI: Differentiated English Teaching and Cultural Competence in Junior Middle Schools

摘要

The implementation of stratified English teaching in junior middle schools consistently encounters two fundamental challenges: the persistence of static "pseudo-stratification" that fails to adapt to students' evolving needs, and growing ethical concerns regarding intrusive artificial intelligence technologies in educational settings. In direct response to these dual challenges, this study proposes and empirically validates an innovative ethics-first lightweight AI framework specifically designed to enable truly dynamic and culturally responsive differentiated instruction. By strategically integrating privacy-preserving federated learning methodologies with non-intrusive mobile sensing technologies, our framework introduces a novel pedagogical architecture termed the "three-layer dynamic grouping and four-dimensional strategy matrix." To comprehensively evaluate its efficacy, we conducted a rigorous six-month controlled experiment involving 326 students across diverse socioeconomic contexts—including urban (eastern), county-town (central), and rural (western) regions of China. The empirical results demonstrate the framework's remarkable effectiveness across multiple dimensions: it achieved a stratification accuracy of 89%, significantly reduced grammatical error rates by 31%, and enhanced cultural understanding by 44%. Most notably, the implementation of a carefully designed tiered informed consent protocol successfully reduced parental data withdrawal requests to an unprecedented 0.9%. By leveraging ubiquitous smartphone sensors for low-cost deployment, this framework provides a scalable, ethically-grounded solution for implementing sophisticated differentiated instruction in resource-constrained educational environments, thereby advancing the crucial goal of educational equity from theoretical principle to practical reality.